fix(inverter): stop spurious "REST failed to setExportTarget" warnings

[springfall2008]

Fixes #4404

Problem

Since v8.47.1, GivEnergy REST (GivTCP) users get Warn: Inverter 0 REST failed to setExportTarget on every export slot, and the run ends Exporting with Errors reported. @davemilsom2 bisected it to 8.47.1 and pointed at #4384, which is right — that PR is the only inverter.py change in v8.47.0..v8.47.1.

There are two defects, one old and one new. They only combine from 8.47.1.

1. rest_setDischargeTarget could never report success (pre-existing)

The write was verified with:

if self.rest_data["raw"]["invertor"]["discharge_target_soc_1"] == target:

GivTCP reports the raw invertor registers as strings. From the attached debug file:

discharge_target_soc_1: '0'

So '4' == 4 is always False. Every call burned all five retries — five redundant register writes and five runAll polls — and then recorded an error, even when the register had taken the value.

The reporter's log proves the write worked:

• 44 × Set export target slot 1 ... via REST failed, 0 successes
• yet 14 × Current discharge target is already set to 4.0 — that path in adjust_force_export coerces with float(), so it sees the value the verify loop refused to match
• 01:50:45 "failed", 01:55:29 "failed", 01:56:13 register reads back 4.0

rest_setReserve immediately above already does this correctly with int(float(...)).

2. 0 was doing double duty as "could not read it" (the regression)

Both branches of the export target block fall back to current = 0 when the value is None or unparseable. While the guard was current > self.reserve_percent that sentinel was harmless — it never triggered a write. #4384 changed it to current != target_soc, so an unreadable register now looks like a target parked below the reserve and gets written every cycle.

That is @russdan's second symptom, on the entity path:

Warn: Inverter 0 write_and_poll_value: Current state for discharge_target_soc is None

Predbat writing to an entity that isn't there.

Fixes

rest_setDischargeTarget — coerce the read back value with int(float(...)) before comparing, guarded for missing/None, and name the value read in the failure message so the next report is diagnosable.

adjust_force_export — an unreadable target stays None instead of collapsing to 0, and is skipped rather than written, on both the REST v3 and entity paths. #4384's bidirectional tracking is untouched for real values. The skip is a plain log line, not a Warn: — an inverter that simply has no such register would otherwise print a warning every cycle.

INVERTER_REST_TIMEOUT 5 → 10 seconds, to give a busy GivTCP more room to respond.

Tests

test_discharge_target_read_back in apps/predbat/tests/test_inverter.py:

• a string read back counts as success, with one write rather than five
• an unreadable REST export target is not written and records no error
• an unreadable export target entity is not written and records no error

Written first and confirmed failing on the pre-fix code (all six assertions), passing after. The test wraps record_status so it asserts only on export target errors, rather than on unrelated fixture noise from the H M format time entities.

The existing test_discharge_target_tracks_reserve from #4384 used ints in raw, which is why it missed this — real GivTCP sends strings.

./run_all --quick and ./run_pre_commit both exit 0.

Not addressed

@russdan also reported REST failed to setDischargeSlot1. That is a different comparison (Timeslots, strings on both sides, no type mismatch) and the debug file failed to upload, so there is no evidence to work from yet. It may be collateral: fix 1 removes four spurious POSTs plus five runAll polls per export cycle, which was hammering GivTCP immediately before the slot write. Worth a retest on this branch before chasing it separately.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Fixes spurious export-target warning noise for GivEnergy GivTCP REST users by making export-target read-back verification type-safe and by avoiding writes when the export-target value is unreadable/missing (e.g., inverter models without that register).

Changes:

• Treat REST read-back values as strings and coerce before comparing so successful export-target writes can be recognised.
• Avoid writing export-target when the current value cannot be read (REST v3 and entity paths), preventing repeated warnings/errors on inverters lacking the register/entity.
• Increase REST request timeout to better tolerate slower/busier local GivTCP instances, and add regression tests for the above behaviours.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
apps/predbat/inverter.py	Fixes REST export-target read-back verification; skips export-target writes when current target is unreadable; improves failure messaging.
apps/predbat/const.py	Increases REST timeout from 5s to 10s to reduce timeouts on busy REST endpoints.
apps/predbat/tests/test_inverter.py	Adds regression test covering string read-back success and “missing/unreadable target” skip behaviour for REST + entity paths.